Bus Network Adjustment Pre-Evaluation Based on Biometric Recognition and Travel Spatio-Temporal Deduction

Abstract

A critical component of bus network adjustment is the accurate prediction of potential risks, such as the likelihood of complaints from passengers. Traditional simulation methods, however, face limitations in identifying passengers and understanding how their travel patterns may change. To address this issue, a pre-evaluation method has been developed, leveraging the spatial distribution of bus networks and the spatio-temporal behavior of passengers. The method includes stage of travel demand analysis, accessible path set calculation, passenger assignment, and evaluation of key indicators. First, we explore the actual passengers’ origin and destination (OD) stop from bus card (or passenger Code) payment data and biometric recognition data, with the OD as one of the main input parameters. Second, a digital bus network model is constructed to represent the logical and spatial relationships between routes and stops. Upon inputting bus line adjustment parameters, these relationships allow for the precise and automatic identification of the affected areas, as well as the calculation of accessible paths of each OD pair. Third, the factors influencing passengers’ path selection are analyzed, and a predictive model is built to estimate post-adjustment path choices. A genetic algorithm is employed to optimize the model’s weights. Finally, various metrics, such as changes in travel routes and ride times, are analyzed by integrating passenger profiles. The proposed method was tested on the case of the Guangzhou 543 route adjustment. Results show that the accuracy of the number of predicted trips after adjustment is 89.6%, and the predicted flow of each associated bus line is also consistent with the actual situation. The main reason for the error is that the path selection has a certain level of irrationality, which stems from the fact that the proportion of passengers who choose the minimum cost path for direct travel is about 65%, while the proportion of one-transfer passengers is only about 50%. Overall, the proposed algorithm can quantitatively analyze the impact of rigid travel groups, occasional travel groups, elderly groups, and other groups that are prone to making complaints in response to bus line adjustment.

1. Introduction

In recent years, due to the diversion of other modes of transportation, the bus passenger volume has decreased significantly in many Chinese cities. Optimizing bus networks to adapt to this change in the structure of people’s travel is the most important means of ensuring an adequate level of public transportation services. For the management of public transportation, the risk of complaints from passengers is an important factor for determining whether the line adjustment schemes can actually be implemented, especially given that the rigid passenger group and other special groups are often associated with a high incidence of complaints and, therefore, must be considered more carefully.

At present, the bus network adjustment evaluation mainly relies on the macro simulation software of the ‘four-stage’ traditional model [1] method, which requires large-scale manual modeling, estimation of travel demand, and a comparative calculation of evaluation results. This often lacks a refined analysis of different passenger groups, and it is also an inconvenience for system integration. In addition, with the popularization of Big Data applications in public transportation, bus OD inference algorithms have become increasingly mature. The accuracy of OD inference is also constantly improving, along with breakthroughs in biometric technologies, such as fingerprint recognition, facial recognition, and palm print recognition. The OD inference algorithm also plays a crucial role in route optimization, facilitating data-based determinations for route validation and execution [2]. This includes Deng Hongxing [3] and others who proposed a bus OD calculation method based on bus IC cards and GPS data. Wei Qingbo [4] further studied real-time OD inference, which provides an abundant data source for the preliminary evaluation of network optimization. There are also other model methods for predicting behavior characteristics. Random forest has been used to predict whether someone will ride the bus frequently based on their behavioral characteristics [5]. Yulong Pei et al. [6] proposed a hybrid-strategy bus passenger-flow prediction model to focus on the impact of the most important features of the data according to the prediction results. Chen et al. [7] designed a framework for predicting traffic flow in urban road networks using deep learning. Nagaraj et al. [8] developed a deep learning-based method with long short-term memory (LSTM), recurrent neural networks, and greedy hierarchical algorithms for predicting bus passenger flows, while Du et al. [9] designed an LSTM network model for predicting how people move through urban traffic. There are also some studies [10,11,12] on hybrid methods that combine the features of different models to enhance prediction accuracy, with Han et al. [13] creating a hybrid optimization of the LSTM model to predict the flow of bus passengers, Glisovic et al. [14] demonstrating a composite passenger prediction model utilizing the genetic algorithm and artificial neural networks, Xu et al. [15] presenting a road-traffic-flow prediction method based on the autoregressive integrated moving average (ARIMA) and the Kalman filter, and Li et al. [16] combining empirical mode decomposition, sample entropy, and the kernel extreme learning machine to predict short-term flows of bus passengers.

For evaluation research, Wang Zhenbao [17] utilized the fuzzy comprehensive evaluation method for bus line adjustment schemes, while Wei Changhai [18] proposed the line adjustment method based on passenger flow and the quantitative evaluation idea of the impact of passenger flow. Meanwhile, Chen Mian et al. [19] developed methods for identifying alternative stops and OD pairs to assess the impact of passenger flows during route adjustments, while Igor Dakicet et al. [20] investigated the effects of demand intensity, user behavior, and trip length on optimal bus network configurations, and Junyong Jang et al. [21] proposed the bus route sketch (BRS) methodology to identify optimal alternatives from all possible driving route alternatives between a given origin and destination. Additionally, several review papers have focused on optimizing network configurations, considering factors such as operational costs, user time costs, transfers, social welfare, and operator profits [22,23,24].

Generally speaking, while these studies have explored bus route adjustment analysis to a certain extent, they still require further exploration regarding quick calculations (systematizations) and refined analysis of the influence of different passenger groups. Given this background, research on the impact assessment technology of bus network adjustments from a micro perspective is conducted using the universal spatial relationship model instead of the traditional simulation model. This research aims to establish a systematic evaluation process that quantifies the effects of bus line adjustments, providing multi-faceted data support for public transportation management and offering meaningful insights for passengers.

2. Technical Roadmap

The most direct way of quantifying the scale and extent of the affected passenger groups after bus line adjustments at the microscopic is to compare the path change of entire passenger OD pairs. In large cities such as Guangzhou, however, there are over 7000 bus stops, and the calculation times of available paths for an OD pair can be up to tens of thousands of iterations. This is extremely resource-intensive for full-domain path calculations. To address this challenge, it is necessary to reduce the calculation range as much as possible. The three key points for achieving the calculation of bus network adjustment pre-evaluation are the following: ‘Comprehensiveness of path acquisition’, ‘accuracy of passenger allocation’, and ‘guarantee of computational efficiency’.

To begin this process, we first design the input parameters and construct a universal geographic information model of the bus line network that can identify all bus stops affected by line adjustment and then calculate the set of available paths for affected areas both before and after the line adjustment. Next, we store the model, affected stops, and paths in the database.

Following this, an OD inference algorithm is developed, integrating ‘passenger card (code) data + biometric recognition data’ to improve the accuracy of the OD inference. Using this algorithm, the OD stops, lines, and direction of each passenger’s trip can be inferred. These results are then used as input for bus travel demand analysis.

Next, we assign passengers to the new network by comprehensively utilizing actual travel demand, available travel path sets, and path selection models. Finally, the changes in the number of trips, the timeliness, and various other aspects before and after the line adjustment are analyzed from the perspective of individual passengers. Following this, the relevant group analysis can be conducted by integrating the results with passenger profiles. The technical route is shown below in Figure 1.

3. Bus Network Model

The network model should meet requirements in terms of identifying the influence range of line adjustments, the path calculations, the route selections, and so on. One can assume that the line adjustment will only affect passengers on the ‘associated’ lines, including logical and spatial related lines, so the key to designing a bus network model is to reflect the spatial and logical relationship (such as subordination and transfer) between different stops, as well as the relationship between stops and lines. The bus sub-network model can be defined as the integration of stops, routes, and transfer information with all lines that intersect or overlap with the adjusted lines. Classified by elements, this model can be divided into a sub-stop model, route model, transfer model, and various other levels, as shown below in Figure 2.

• The sub-stop model is the smallest unit in a bus network and is used to describe a single line and unidirectional stop. It is also the core model of the entire network model. Its elements include the line that the stop belongs to, the direction (up/down), name, ID, stop order, distance and travel time from the previous stop, latitude and longitude, adjacent sub-stop collection (which refers to the stops set that can be reached by walking), and stop class (which is divided into four classes: Class1, directly adjusted stops; Class2: walkable stops from direct stops; Class3: stops on the same line as direct stops; Class4: other associated stops). The sub-stop ID is unique, which means that different bus lines have different IDs when they pass through bus stops with the same physical properties.
• The route model is used to describe the information about stops upstream and downstream. Such information mainly includes the route name, route ID, collection of upstream and downstream stops, the collection of transfers of each stop (adjacent sub-stops), and the starting and ending overlap stop order between the current line and the adjusted line. The route model is used to calculate the association of the stops on the same lines.
• The transfer model focuses on transfer stop and route information, expressing the relationships between different lines. It records details about the upstream and downstream transfers, the starting and target transfer stops, riding distance before and after the transfer, walking distance during the transfer, and the transfer stop itself.

To automatically identify the impact range of route adjustments, the process begins with obtaining the location information of the stop to be adjusted. Using information such as stop location, route, and walking distance, the system can identify ‘directly related’ (i.e., adjacent, same line) routes and stops. Then, the ‘indirectly related’ stops and routes are identified by utilizing ‘directly related’ stop information, route data, and transfer details. Together, these directly and indirectly related stops and routes form a bus sub-network. Among these factors, ‘walking distance’ is a key parameter that controls the number of ‘related’ stops and routes, making it one of the most important variables in defining the size of the bus sub-network and optimizing the model’s computational efficiency.

4. Pre-Evaluation Based on the Spatio-Temporal Model

4.1. Input Parameters

There are five modes of route adjustments, namely truncation, extension, cancellation, detour, and new opening. These can essentially be classified into two types: the addition or deletion of stops. According to the sub-network model and computational requirements, the input parameters for line adjustment can be specified, as shown in

Table 1. Input parameters.

Data Type		Content of Entry
Line Parameter	Line truncation or cancellation	The ID of all deleted stops
	Extension and Detour or New	Line ID, direction, stop order, distance from the previous stop, line name, ID (unique), stop name, stop type (terminal/midway stop), longitude and latitude, distance from the next stop
Passenger Travel Demand		Passenger ID, passenger type (elderly, students, working), trading time, boarding time, boarding stop, alighting time, alighting stop, travel route, direction, travel distance, etc.

. When stops are deleted, only the ID numbers of the removed stops need to be entered. In contrast, when adding new stops, more detailed information is required, including the line ID, direction, stop order, unique stop ID, stop name, distance from the previous stop, distance to the next stop, and the longitude and latitude of each newly added stop.

After that, based on the information regarding the deleted stops’ IDs and the added stops’ position, each stop’s adjacent sub-stops are identified, and a bus sub-network model is constructed representing two scenes (before and after line adjustment) by clarifying all OD stops affected, including directly adjusted stops, walkable stops among directly adjusted stops, stops on the same line as direct stops, and all stops on all lines that intersect with the adjusted line (walkable transfers). It is important to ensure consistency in the scope of analysis between the old and new sub-networks, particularly when considering ‘deleted stops in the new sub-network’ and ‘added stops in the old sub-network’. This ensures that spatial location searches remain consistent and accurate, avoiding discrepancies in the comparison of paths between the old and new networks. Such inconsistencies could skew the results of subsequent analyses.

4.2. Passenger Travel Demand Analysis Based on Card (Code) Data and Biometric Recognition Data

In the adjustment of public transportation networks, passengers’ travel demand is manifested as the origin and destination (OD) stops, travel volume, and travel time distribution. With the traditional ‘four-stage method’, passengers’ travel needs are obtained based on manual surveys or statistical data sets. This article comprehensively uses passenger card (code) data, biometric-based payment data, and bus operation data (operating time, vehicle location, and operating route) to analyze the travel information of each passenger in the current bus network. The main steps for doing this include matching boarding stations, inferring alighting stations, and identifying the true origin and destination stops of transfers.Process of passenger travel demand analysis is shown below in Figure 3.

At present, public transportation generally adopts a ‘one ticket’ payment method—passengers only need to swipe or scan when getting on the bus, without doing so when alighting. This makes it relatively straightforward to identify the boarding stop by associating and matching payment information with bus operation data based on time. To infer the alighting stop, the general principle is as follows: first, define the ‘set of possible alighting stops for this trip’, $D$—the set of all downstream stations on the route they are traveling on. Second, define the ‘set of surrounding stations the passenger could next board’, $O$—the set of next boarding stations and their surrounding stations (within a radius of 0.5–1 km). Third, if $O\cap D=Q\ne \varnothing$, the nearest station to the next boarding station at a medium distance $Q$ will be used as the station of alighting for this trip. After obtaining the destination stops, the transfer behavior can be determined based on the time difference between the two trips (the next alighting time—the current boarding time ≤ 30 min), and the true origin and destination stops of the passenger’s trip can be obtained.

In addition to card and code data, face and palm print recognition plays a significant role in modern public transportation. At present, this type of recognition technology is being promoted and applied in public transportation payment scenarios in various cities. In this study, a system is proposed to collect real-time facial and palm print data (with passenger consent), which is then cross-referenced with a historical biometric database to identify passengers. Combined with passengers’ electronic wallets, this facilitates biometric-based payments. Multimodal convolutional neural network biometric fusion was adopted to improve recognition efficiency and accuracy [25]. The public transportation demand analysis mainly looks at passenger identity information, travel payment time, payment terminal information, and payment amount. Passenger identity information, payment times, terminal data, and payment amounts are analyzed to provide a comprehensive picture of travel demand. This approach integrates biometric data with card/code payment data, offering advantages for more detailed analyses, including multimodal travel patterns.

By utilizing this method, the origin and destination stops, travel times, and travel volumes of each trip within the current bus network can be inferred. Moreover, biometric and card data offer additional identity insights, forming a robust data foundation for further detailed analyses.

4.3. Accessible Path Analysis Between Origin and Destination

Bus accessibility between an origin and destination (OD) can be classified into three categories: direct, one-transfer accessible, and inaccessible. For bus transportation, routes requiring more than two transfers are often less attractive and can be deemed as inaccessible. The calculation of accessible paths between the origin and destination is as follows:

4.3.1. Direct Path Analysis

The key to direct path analysis is to obtain the line passing through the origin and destination stops and obtain all the stopping lines that can be reached by walking at both stops (including themselves). If there are lines passing through both the origin and destination, the ODs are considered directly reachable. According to the line model, direct path information is obtained, including stop-to-stop distance, walking distance, travel time, line, and direction.

4.3.2. Primary Transfer Path Analysis

For an OD without direct lines, its core is the ‘transfer line’ analysis of all downstream stops from the origin stop. First, we find all the passing lines of the origin and destination stops through stop names and walking distances. Second, we find all downstream stops for the origin stop from the line model according to the lines, stop order, and direction. Third, we obtain the transfer line collection $L_f$ of all downstream stops according to the transfer model of each downstream stop. Finally, we compare the transfer line collection $L_f$ and the stopping line collection $L_d$ of the terminal stop; if $L_f\cap L_d\ne \varnothing$, it means there is a single transfer path between the origin and the destination. We record the starting and ending stop, direction, distance, travel time, and transfer walking distance (space) of the upstream and downstream lines.

For example, assuming that ‘Route L’ in Figure 4 above is the adjusted line and S is the associated line, the passenger travel between the intersecting downstream lines, such as stops $S_1$ and $S_2$ in the associated lines, is not affected by the adjustment of Route L, and the paths of this type of OD do not need to be calculated.

To further improve computational efficiency, only paths between OD stops within Class 1–3 (in Chapter 3) or OD stops between Class 1–3 and Class 4 are considered. This reduces unnecessary calculations and storage for paths between less relevant stop pairs, particularly those within Class 4 stops. This selective calculation helps streamline the overall analysis and ensures faster processing of bus network adjustments.

4.4. Passenger Path Selection Model

After calculating the OD accessible path set, the next step is to assign passenger travel demand to the most ‘optimal’ path. The path selection of passengers is influenced by various factors, which need to be comprehensively analyzed based on available data.

These factors typically include the following: (1) passenger attributes: age and occasional or rigid travel; (2) path attributes: distance, whether or not transfers are necessary, the number of stops, the waiting time, and the walking time; (3) passenger travel habits: whether a travel route is habitual and the frequency of trips taken; (4) the relationship with other paths: the number of replaceable paths and the ratio to the optimal path.

4.4.1. Analysis of Influencing Factors

Analysis of Passengers’ Attributes

Passenger demographics and behavior significantly affect route selection. For example, by analyzing composition of Bus Passenger Travel in Guangzhou, as shown in Figure 5, we found that the elderly, disabled, and other passenger groups who enjoy high discounts have lower time sensitivity, and the proportion who choose transfer routes is close to or exceeds 20%. In contrast, the working group and student group prioritize timeliness, with less than 10% opting for transfer routes. In addition, the ratio of trips with transfers during off-peak periods slightly increases by two percentage points. These findings underscore the impact of passenger attributes, including travel time, on route selection.

Analysis of Passenger Travel Habits

By tracking the frequency of trips on the same OD pair over a week, the most popular route for each passenger can be identified. The proportion of the most popular bus routes in the same OD travel is shown in Figure 6. From that, one can see that bus passengers tended to take a fixed route, with stronger preferences as trip frequency decreased. When the weekly travel frequency of the same OD is three times, the proportion of passengers who choose the same route is close to 90%. As the weekly travel frequency increases, this preference gradually decreases, but overall, it exceeds 50%. This indicates that passengers generally prefer familiar paths when selecting a route.

Analysis of the Impact on Travel Efficiency

Several direct factors influence route choice, including ‘travel time, walking time, and waiting time’. By integrating data on passenger’s travel OD (see Section 4.2) and accessible paths (see Section 4.3), the relative importance of each factor in passengers’ path selection can be analyzed. The direct efficiency function is shown in Formula (1) below:

$$E=T_{route}+T_{wait}+T_{walk}$$

(1)

in this formula $E$—is the efficiency. T_route—represents the travel time, T_wait—represents the waiting time, and T_walk—represents walking time (including walking to/from the stops and during transfers).

Alternative Path Impact Analysis

Ideally, passengers would choose the optimal travel route, but given that they are often limited by incomplete access to information, differences in personal habits, and other factors, the path choices of passengers for transfer travelers, in particular, are not entirely rational and feature noticeable diversity. For example, Ma GeHui [26] found that the total time difference factors have the most significant impact on passenger path selection, while Chen Mingli [27] found that bus passenger path decisions are influenced by both selfishness and rationality.

For all possible paths on a given OD pair, the efficiency $E$ is calculated for each path and compared to the optimal path. The path selection distribution is shown below in Figure 7, which indicates that the majority of passengers prefer the ‘optimal path’, whether for direct travel or transfer travel, with proportions reaching 65% and 50%, respectively. As $E$ increases, the likelihood of selection decreases, though the probability remains non-zero. Meanwhile, the distribution curve of different travel periods, such as peak and off-peak, basically coincides. In the case of transfers, passengers are less likely to take the ‘optimal’ path than they are for direct travel. This is mainly due to factors such as the integrity of information acquisition.

According to the curve in Figure 7 and the relationship between the path selection probability R and the ‘optimal path efficiency ratio’ of direct and one-time transfer paths, Formulas (2) and (3) are developed as shown:

$$R=\left\{\begin{array}{ll}0.65\hfill & \hfill t^{\prime }=1\\ 0.012\times {t^{\prime }}^{-3.868}\hfill & \hfill t^{\prime }>1\\ t^{\prime }={\displaystyle \frac{C}{C_{\min }}}\hfill \end{array}\right.$$

(2)

$$R=\left\{\begin{array}{ll}0.486\hfill & \hfill t^{\prime }=1\\ 0.07\times {t^{\prime }}^{-8}\hfill & \hfill t^{\prime }>1\\ t^{\prime }={\displaystyle \frac{C}{C_{\min }}}\hfill \end{array}\right.$$

(3)

4.4.2. Path Selection Function

Path Selection Function

Based on the above analysis, the state vector for path selection can be set as:

P = [ID, G, O, D, T_o, T_route, T_wait, T_walk, P_num, H, H_num, R, A]

where:

ID—passenger ID;

G—group type;

O—origin bus stop;

D—destination bus stop;

T_o—departure period (such as morning peak, evening peak, off-peak, or nighttime);

P_num—the number of stops passed;

H—whether the route taken is a habitual travel route, represented by 0 or 1;

H_num—historical travel frequency;

R—optimal path time ratio;

A—number of replaceable paths.

ID, G, O, D, and H are mainly used for matching relevant elements in travel history records. In specific passenger travel assignments, if these new travel elements are consistent with historical travel, they are taken as 0, otherwise, they are taken as 1.

Based on the construction of the vectors above, according to the requirements of reliable, simple, and traceable engineering applications, a passenger path selection index is set as Formula (4). The smaller the value of the $\delta$, the higher the probability of that path being selected:

$$\begin{array}{l}\delta ={\alpha }_{1}O+{\alpha }_{2}D+{\alpha }_3{T}_o+{\alpha }_4{T}_{route}+{\alpha }_5{T}_{wait}\\ \hspace{1em}\hspace{1em}+{\alpha }_6{T}_{walk}+{\alpha }_7{P}_{num}+{\alpha }_{8}H+{\alpha }_9{H}_{num}+{\alpha }_{10}R+{\alpha }_{11}A\end{array}$$

(4)

Weight Determination

To obtain the optimal weight combination in Formula (4), the genetic algorithm is employed. The process includes individual mathematical coding design (real number coding), initial population generation (hundreds of individuals), fitness function design (mathematical description of optimization objectives), crossover and mutation selection (survival of the fittest), and optimal individual selection. Of these, fitness function design is the most critical step. The fitness function is ‘the maximum accuracy of the routes assigned to all bus passengers’ and is derived from the actual travel OD and route inferred based on the analysis of ‘card (code) + biometric’ data. The formula of the fitness function is as follows:

$$S={\displaystyle \sum _i^n{S}_i},\hspace{1em}\hspace{1em}{S}_i=\left\{\begin{array}{ll}0\hfill & \hspace{1em}\hspace{1em}\hfill {{\mathrm{L}}^{\prime }}_f\ne {\mathrm{L}}_a,\hspace{1em}\mathrm{O}{{\mathrm{D}}^{\prime }}_f\ne \mathrm{O}{\mathrm{D}}_a\\ 0.5\hfill & \hspace{1em}\hspace{1em}\hfill {{\mathrm{L}}^{\prime }}_f={\mathrm{L}}_a,\hspace{1em}\mathrm{O}{{\mathrm{D}}^{\prime }}_f\ne \mathrm{O}{\mathrm{D}}_a\\ 1\hfill & \hspace{1em}\hspace{1em}\hfill {{\mathrm{L}}^{\prime }}_f={\mathrm{L}}_a,\hspace{1em}\mathrm{O}{{\mathrm{D}}^{\prime }}_f=\mathrm{O}{\mathrm{D}}_a\end{array}\right.$$

(5)

where S—the accuracy of passenger assignment; n—the number of passengers in the travel group; $S_i$—the accuracy of one passenger’s path: When the path (bus line and OD stop) assigned to passenger i is exactly the same as the actual route, the value is 1. When the line is the same, but the OD stop is different, the value is 0.5; in all other scenarios, the value is 0.

Different passenger groups and travel time periods require different weight combinations, which are optimized through GA. The results are shown in

Table 2. Weight coefficient.

Group	Time Period	Weight Coefficient
		α1	α2	α3	α4	α5	α6	α7	α8	α9	α10	α11
working	Peak	91	80	70	3	5	13	1	33	40	64	44
	Off-peak	97	90	31	1	2	8	1	6	10	19	34
elderly	Peak	87	88	21	9	12	67	1	4	8	2	51
	Off-peak	56	97	46	2	13	37	3	1	24	8	93
student	Peak	81	80	30	3	5	25	1	8	77	8	25
	Off-peak	97	76	7	19	12	70	5	14	30	50	35
disabled	Peak	87	97	73	4	28	94	3	27	7	24	29
	Off-peak	39	82	75	10	1	50	1	68	31	11	35
other	Peak	99	85	91	7	4	59	42	49	50	59	85
	Off-peak	77	97	24	11	3	40	4	6	40	14	22

.

The weights of different factors reflect the tendency of passengers in the path selection process in a certain program. Table 2 shows that the consistency of the origin and destination stops is the most important factor (α₁,α₂). Other notable factors include the travel time period (α₃), walking time (α₆), whether a path is a usual travel path (α₈), historical frequency (α₉), optimal path efficiency ratio (α₁₀), and the number of replaceable paths (α₁₁). The coefficients of travel time (α₄), waiting time(α₅), and the number of stops (α₇) passed through are relatively small. This reflects that passengers consider more comparisons when choosing rather than just the route itself.

4.5. Passenger Path Comparison

The path with the smallest $\delta$ is used as the optimal selection path, while a random method is used to determine multiple paths when $\delta$ values are completely equal. Compared with the original travel route, one can analyze the changes in transfer number, distance, and travel time of each passenger after a route adjustment. The passenger section changes of an affected line can also be compared. The specific process is shown in Figure 8 below.

The key for passenger path comparison is the real OD identification of passengers. This mainly includes transit OD analysis, basic data preparation, transfer identification, and original path matching. In engineering applications, one can not infer all passengers’ boarding and alighting stops, so the gravity model method can be used to fix the missing OD. On this basis, one can identify the transfer trip so as to obtain the ‘real OD’ of passengers for a complete trip. The pre-evaluation of line adjustments must be carried out with ‘real OD’ to avoid repetition errors.

5. Case Study Analysis

To validate the proposed methodology, a case study involving the adjustment of Line 543 in Guangzhou is conducted. Utilizing C# and Oracle for system development, we implemented bus model construction, path set calculations, and passenger assignments. The total operation time for this calculation process is about 6 h and meets engineering application timeliness requirements.

Line 543 (highlighted in Figure 9), located in northwestern Guangzhou’s central urban area, features 27 stops and ceased operations on 18 June 2023. A pre-evaluation of passenger travel changes following route adjustments was performed and subsequently verified against real travel data collected on 20 July 2023. Prior to the route’s suspension, the daily passenger traffic volume was about 2290, with the line connected to 109 associated sub-network lines (depicted as gray solid lines in the figure), accounting for one-seventh of the bus lines in the central urban area of Guangzhou. The analysis reveals a complex network of 998 original stops and 978 destination stops within the sub-network, with calculations yielding 106,000 directly related OD pairs and a total of 2.77 million routes, corresponding to 560,000 passenger flows throughout the entire sub-network.

5.1. Calibration of Results

This section will focus on the accuracy of individual passenger path selection. The travel habits of the original passengers changed significantly after the Line 543 adjustment, and there was no travel on July 20. For this reason, only Line 543 passengers who had the same travel requirements before and after the adjustment were sampled for calibration inspection, and their travel changes were analyzed; 395 trips were recorded, accounting for 17% of the total passenger flow for this line. Of these trips, 354 (direct, transfer) were correctly judged, with an accuracy rate of 89.6%. The two main sources of errors were in the OD inference process and in what different passengers deemed to be acceptable walking distances.

The passenger flow transfers between Line 543 and other lines were also examined. Passengers mainly transfer to five other routes, notably Line 238 and 539, with more than 150 transfer passengers recorded daily.

Table 3. Passenger flow changes in main lines (trips/day).

Bus Line	Actual Flow Before Adjustment	Actual Flow After Adjustment	Estimated Flow	Relative Error	Difference Between Actual And Estimated
Line 238	4491	4759	4759	0.00%	268
Line 539	5723	6052	5992	−0.99%	269
Line 215	12,579	12,600	12,842	1.92%	263
Line 556	5690	6262	5898	−5.81%	208
Line 524	4708	4980	4867	−2.27%	159

presents a comparison of line passenger flow for both pre-evaluation and actual scenarios (20 July 2023). The results indicate a close match, with a relative difference of less than 10%. Due to the inherent randomness of travel behavior, the actual passenger groups for these lines may vary before and after the adjustment; thus, the relative difference should be viewed as a numerical reference rather than a definitive error. This analysis suggests that the proposed method is effective in predicting the scale of passenger transportation after the adjustment.

5.2. Pre-Evaluation Analysis

The pre-evaluation indicator system includes the transfer of passenger flows, the scale of affected passengers, the changes in the number of rides, the passenger profile, and the change in a line’s passenger section volume. The detailed indicators are shown in

Table 4. Pre-evaluation key indicators.

	Passengers Flow	Rigid Passengers
Trips with more than 2 transfers	0	0
Trips with increasing riding numbers	645	328
Of which are elderly	128	35
Of which are students	14	9
Of which are disabled	5	4

. Overall, the adjustment of Line 543 will not result in an increased number of unreachable passengers (including those who require more than two transfers to travel to their destinations), but there are 645 trips where the number of transfers increased by one accounting for 28% of all the passengers on Line 543. Of these trips, 328 of them (a total of 281 people) could be classified as rigid travel. Meanwhile, the affected flow of vulnerable groups such as the elderly, students, and the disabled were 128 trips, 14 trips, and 5 trips, respectively. There is also a certain rigidity in the overall travel habits, as they are more prone to complaints and require careful consideration.

In addition, by randomly analyzing the change in section passenger volume in Line 238, as can be seen in Figure 10, the estimation results and the actual results are quite similar. This means that the data can also meet the requirements of capacity allocation calculation in bus operation.

6. Conclusions

This paper establishes a pre-evaluation method for bus network adjustments that moves beyond traditional commercial simulation software. By conducting in-depth research on critical steps and methodologies based on Big Data mining, the study enables a refined analysis of changes in ridership among different passenger groups and variations in passenger volume across various line sections. The main conclusions drawn from this research are as follows:

• The calculation of accessible paths is crucial for effective pre-evaluation. To enhance calculation efficiency, the core mission is to design a bus network model, recording information such as stops and bus lines accessible via walking or transferring. This involves controlling the calculative size through the affected bus sub-networks. Additionally, employing the ‘stop class’ method can significantly reduce unnecessary path calculations for unrelated OD pairs, thereby decreasing computational complexity.
• Accurate bus passenger flow distribution is essential. In the case of multiple paths, passenger travel choices are not always completely rational. Although these choices generally conform to the principle of ‘the higher the path cost, the lower the probability of being selected’, a degree of randomness is also evident. The proportions of direct and optimal paths are 65% and 50%, respectively. The selection probability curves of different travel times—such as peak and weekday—are nearly the same.
• Passengers tend to consider more comparisons rather than focusing solely on a single route. The main factors they consider include consistency between the origin and destination stop, walking time, whether a route is familiar, historical travel frequency, the efficiency ratio between the current and optimal path, and the number of alternative paths.
• According to the tracking of passenger samples before and after adjustment, the proposed method demonstrates an accuracy rate of 89.6% in estimating changes in trips (both direct and transfer). Moreover, changes in passenger flows and line section traffic volume closely align with actual observed outcomes.
• The primary reasons for evaluation errors are related to two aspects: inaccuracies in OD inference and variations in what different passengers consider acceptable walking distances.
• Compared with the conventional four-stage method, this proposed model requires only the relevant parameters of the line being adjusted. This model enables the automatic calculation of evaluation results and insights into the varying impacts on different passenger demographics. This represents a certain degree of improvement in terms of both evaluation efficiency and refinement.